Various methods are provided for enabling the application of machine learning to network management and in particular to enabling cognitive network management in radio access networks. One example method may comprise interpreting one or more operator goals for the CNM or for a specific CF to ensure that the specific CF adjusts its behavior in order to fulfil the operator goals, abstracting an environment into states configured for use in subsequent decision making, wherein the abstracted environment represent are built from one or more of a combination of quantitative KPIs, abstract state labels, and operational contexts, defining legal candidate network configurations for different contexts of the CF based on the abstracted environments and operational contexts as inferred by the EMA engine, and matching a current abstract state, abstracted environment, or operational context as derived by the EMA engine to an appropriate network configuration selected from the set of legal candidate network configurations.